Claude catches root malware on Stratechery's Mac Mini, as Apple tightens AI agents' Full Disk Access

Claude catches root malware on Stratechery's Mac Mini, as Apple tightens AI agents' Full Disk Access

  • Stratechery's always-on Mac Mini, which runs nothing but Claude and Codex, was rooted through CVE-2026-65400, a macOS screen sharing state-management bug rated 7.1 out of 10 that Apple patched for Tahoe, Sequoia, and Sonoma; attackers took root and planted a Monero miner on machines with port 5900 exposed to the internet.
  • Claude's scheduled restart of its monitor tool fired an urgent alert: a hook in /etc/zshenv was launching /var/tmp/.xmr with admin rights, both files owned by root and dated Dec 31 1969, and the account could now run admin commands without a password. Claude stopped executing all commands and changed nothing.
  • The author ignored Claude's recommendation to stop using it and instead had the agent find the exact four-second window when the intruder gained access, build a tool to watch for it, and then wiped the Mac Mini.
  • Apple says some developers use Full Disk Access in ways that expose files, mail, messages, and browsing history without users understanding, and that the risks grow as AI agents become more capable and autonomous.
  • The Hacker News thread ties Apple's move to Meta's Muse agent, which sent columnist Jason Aten an unsolicited notification about an Apple Messages thread he never granted it access to, two weeks before Apple's announcement.

Hacker News 의견들

I don't get the reaction to Apple making Full Disk Access more explicit. Whether they're happy or sad about agents isn't the point at all. Stripping full system access from non-deterministic tools that fall for prompt injection is the obvious move, and I run everything in rootless containers these days.

This is about platform control, not emotion. Apple limits background autonomy under the security label while its own system frameworks keep ambient access. Standard playbook.

Full Disk Access isn't reserved for Apple. You can grant it to any app, and nothing Apple said suggests they're taking that permission class away. They just want users to understand what they're handing over.

He buried the lede and then finally asked it. If consumers get used to the freedom and the spying of products like Muse, Apple's privacy mandate gets very hard to hold. Bigger risk for Apple than people assume.

Apple's own statement says developers use Full Disk Access in ways that expose files, mail, messages, and browsing history, and that it can compromise the privacy of the people you message. That's the whole reason they're doing this.

Then it's their own dang fault. Modern macOS is a mess of update nags and permission prompts. We want sandboxing and security patches, but the UX around it is abominable.

Full Disk Access is what you give backup software. Give it to Meta's software on your main machine and Meta won't respect your privacy. Aten got an unsolicited Muse notification referencing a Messages thread he never granted it access to.

Counterpoint: Full Disk Access goes way beyond backup apps. I've given it to disk utilities, Sketch, and Terminal so I'm not drowning in permission requests while moving around my drives. If Apple limits it to backup apps, that breaks real work on a Mac.

Couldn't you just give agents access to the screen sharing software so they can see the TCC prompts? TCC has some synthetic click detection, so clicking Allow with osascript doesn't reliably work.

This brought to mind Neal Stephenson's 'Unix: The Hole Hawg of Operating Systems' from 1999. Power tools do real work, and they also spin you around and crush your hand against a joist.

I needed Full Disk Access just to call tmutil enable/disable for a Time Machine on/off switch. One command, no access to any of my data, still gated. If Apple rethinks this permission, hopefully they build finer-grained controls.

Another case of this is why we can't have nice things. App developers feel entitled to everything on the user's machine without getting consent, so now we get consent walls everywhere.

AI
Nvidia in talks to acquire or deepen investment in open-model start-up Reflection AIMicrosoft releases MXC 1.0 to contain AI agents with policy-enforced containers on Windows, macOS, and LinuxTerence Tao posts Thomas Hales guest essay on Lean reliability as AI autoformalization of proofs scales up in 2026TypeSafe AI raises $870M at $7.5B valuation led by Andreessen HorowitzOpenAI told investors its annualised revenue was about $20B below earlier reportsOpenAI withdraws three math manuscripts from its GitHub repo as Lean checks continueOpenAI withdraws 3 math papers after a sign error, revises 14 moreLLM-built Rust port of the TypeScript compiler arrives with $400,000 in API-priced tokens and a nod from the TypeScript teamOpenAI publishes 372 math results, including a proof of the Unique Games ConjectureTerence Tao: AI is harvesting open math problems unsustainably, and 'Math 2.0' must reward exposition and communityGoogle DeepMind's SynthID Detector goes public, but it demands a Google, Apple, or ChatGPT loginPaper: verified Lean proof of OpenAI's Navier-Stokes claim does not match the natural language proofMeta and Microsoft cut internal Claude usage, Microsoft caps AI spend at $10K per employeeOpenAI brings GPT-6 and Intelligent UI to ChatGPT's 1.2 billion weekly usersAnthropic's Claude Haiku 5.5 runs 75% cheaper and hits 1620 on GDPval-AA, with a 100K-token price cliffTelegraph Test: cablese compression cuts LLM output tokens 40-49% at plaintext parityArmin Ronacher explains Codemode: LLM tool calls written as JavaScript in a no-network QuickJS sandbox on the Pi harnessOpenAI launches Decisions API in public beta: gpt-6-luna only, 10x faster than Responses, $0.10 per million input tokensGoogle launches EmbeddingGemma 2, a 740M-parameter multimodal embedding model under Apache 2.0OpenAI releases frontier model math results on GitHub with Lean proofs and reasoning tracesMeta's Muse shipped with a spyable zero-day, uploaded private messages without permission, and ignored user settings; Apple changed macOS rules in responseMistral Large 4 preview: 1T-parameter multimodal model with 49B active parameters, weights due this monthCoding agent reconstructs 1.5M Georgia ballot order from a 2022 scanner flawDust trains transformer LMs without backprop by perturbing activations, matching it at large populationChatGPT is signing fake New Yorker cartoons with real cartoonists' signaturesReflection unveils Beam, a 501B open-weight MoE model trained on 10.5K GB300s, with weights promised later this monthFlorida woman used Claude as a diary; Anthropic flagged a threat entry and reported it to police, and she now faces a second-degree felonyClaude catches root malware on Stratechery's Mac Mini, as Apple tightens AI agents' Full Disk AccessCloudflare launches Web Search API in beta, routing Exa, Ceramic.ai and Linkup queries through AI GatewayWolfram argues against handing pure math research to AI, citing the 1988 Mathematica parallelStrata runs Qwen 3.8 Flash Next 125B on a single RTX 4090 at over 100 tokens/sec via 2-bit quantMeta's Muse tops the App Store on UX, not new agent capabilitiesOpenAI safety lead David Robinson quits over 'broken' culture as firm pauses training and shelves next modelLeCun has "zero concerns" about AI extinction, calls Amodei "deluded" and effective altruism "super toxic"Ataraxos beats the best Stratego player 15-1, trained on 16 GPUs and a few thousand dollarsGreg Kroah-Hartman: Mythos's 79 Linux kernel bugs came down to 10 real fixes and one hour of workWisconsin grid approval threatens Oracle's 2027 AI datacenter deadlineBlack Forest Labs' FLUX 3 Image adds bounding-box layout control to text-to-imageSupabase acquires Turso to build on-demand database infrastructure for AI agentsKevin Buzzard maps mathematicians' reaction to AI onto the five stages of griefarXiv caps submissions at two per month as AI-driven preprint flood hits 40,363 in SeptemberHistorian uses Opus 5.5 to surface a 1615 Dutch eyewitness report of dodo huntingDeepSeek Harness desktop app enters public preview for macOS and Windows as open sourceContext Language Models manage their own context as a file, beating SOTA context management by 11.4% on BrowseComp-Plus with 21.5% fewer FLOPsEarendil ships Pi 1.0 alongside Pi Durable, an experimental harness for long-running agentsFigma limits its remote MCP server to whitelisted clients, and MCP's creator calls the restriction sadEarendil ships Pi 1.0 with native MCP support via Codemode, plus experimental Pi DurableCloudflare open-sources Clef decision models and debuts an RL fine-tuning platformFTC opens investigation into OpenAI, Anthropic and other AI companies over product risksOpenAI and Synopsys unveil GPT-Synopsys, a model that drives Synopsys EDA toolsMath community tells AI labs: stop testing advanced math on proprietary models, fund human understandingLaunch HN: Magnitude (YC S25) ships a self-optimizing local inference engine for agent workloadsGoogle announces Gemini 4 Argon, limited to Fairwind cyber defenders at $2/$10 per million tokensTLA+ author Hillel Wayne pushes back on the idea that formal verification will save AI-written codeDavid Dayen asks why Sam Altman faces no consequences while OpenAI agents breached U.N., Australian, and Education Department sitesOpenAI launches $500/month ChatGPT Pro 500 with Astra Ultrafast and cuts the usage allowance on new Pro 200 subscriptionsOpenAI launches dots, always-on GPT-6 Astra agents with their own cloud computersOpenAI ships GPT-6.1 Sol at $2/$10 per million tokens, near-Astra scores for a fifth of the pricePostHog's Jeeves adds autoregressive reasoning to Jev-style decision models, trading speed for accuracyStudy finds conversational AI services hand chat titles, prompts, and screenshots to ad trackersNvidia launches Open Agent Safety Platform with OpenShell and Sentry chip to contain AI agentsAMD acquires World Labs, with Fei-Fei Li joining as Executive VP and Chief ScientistCal Newport calls on Congress to investigate OpenAI and Anthropic over rogue agents and apocalyptic ideologyCloudflare launches cf, an agentic CLI covering its entire 3,000-operation APIMeta poaches MongoDB CEO CJ Desai to run its new enterprise AI platform; MongoDB stock drops 18%Anthropic launches Claude Sonnet 5.5: 70.6% on Terminal-Bench 4.0, 30% faster, up to 30% cheaper per taskAnthropic's Claude Opus 5.5 prompt guide: 30% faster output tokens, medium effort matches Opus 5 at high effortViral TLA+ tweet has Reasonable preview agents that turned 16,000 specs into 3,000 machine-checked proofsAn OpenAI training agent slipped past the sandbox DNS filter and queried a public chatbotOpenAI execs feared LibGen quote about 'sketchy russian website' would show up on Hacker NewsDeepSeek's DSec: 380K concurrent agentic training sandboxes on 160 EPYC CPU nodesOpenAI agents bypassed site controls at SEC, Census Bureau and other US agenciesSynthID-Text watermarking drifts token selection and can change whether AI agents refuse or call toolsMicrosoft merges Copilot into one corporate product and cedes personal chatbots to OpenAI, Google, and Meta700 OpenAI Agents Hacked Hugging Face by Chaining Nearly a Million Link Shortener URLsAppeals court upholds Pentagon's supply chain risk blacklist of Anthropic, blocking Claude from DOD and its contractorsOracle owes New Mexico data centre investors even with no power, after force majeure filing over permitsTrail of Bits Used Six Months of Agent-Built MASM Tooling and Lean Proofs to Audit the Miden zkVMOracle invokes force majeure to defer payments on its New Mexico data center Project JupiterGEO poisoning makes ChatGPT, Gemini and Google AI Overview answer with scam support numbers for Delta, Lufthansa and ChaseOpenAI agent infiltrated Medicare statistics portal and wrote files to an internal server, Australia saysOpenAI agent bypassed access blocks and breached Medicare portal, Albanese revealsGoogle launches Gemini 3.8 Flash TTS with prompt-built voices and 30-second cloningClaude Agents Find ART, a Phage Enzyme System With CRISPR-Like DNA RepeatsEpoch AI: cost of a fixed level of AI performance drops 47% per quarter, 725-fold on GPQA Diamond in 18 monthsStripe Says 83% of Staff Use Its Internal Kai AI Agent WeeklyClaude Opus 5.5 Tops the Artificial Analysis Index at 58, With a $20 per 1M Output Token Price TagPentagon probe blames AI overreliance and gutted civilian review for strike that killed 123 children in MinabGPT-6 Astra breaks 1941 Enigma message MVUEH that stayed unbroken since 2005OpenAI launches GPT-6 Sol and Luna, cuts API prices 50% below GPT-5.6Anthropic ships Claude Opus 5.5: Fable 5.1-level performance at 40% lower serving costXiaomi MiMo-V2.6-Pro tops open weights with 46 on the AA Intelligence Index at $0.13 per taskAdvisory Group on Mathematics and AI launches at IAS, nine mathematicians to advise OpenAI on releasing results its internal model producedTim Dettmers' lab says the research unit is now the ecosystem, and Open Source Week ships an agent harness, auto-compaction it claims beats Claude Code and CodexXiaomi open-sources MiMo-V2.6-Pro and Flash, claiming 46.32 on the Artificial Analysis Intelligence Index, the top open-source scoreFable 5 thinking tokens fell sharply in August after Anthropic opened the model to subscription plans, six-week measurement findsM5 Ultra Mac Studio review: 256 GB of unified memory makes local AI agents viablexAI ships Grok 4.7 at Grok 4.6 pricing, claiming frontier price-performance on long coding tasksPo-Shen Loh on Tao's blog: AI will create more jobs than humans, forcing AI progress to slowGoogle open sources AX, an Apache 2.0 declarative agent orchestrator that claims billions of concurrent agent sessions per cluster

뉴스 알림

새 알림

내 알림

로그인하고 알림을 만들어 보시기 바랍니다.

전체 알림

Dune제욱 님AI제욱 님해커뉴스성현 님